A new target tracking algorithm based on muti-domain network is proposed to track the aircraft target in real time. 

Recently, a research team has proposed a fast deep learning for aircraft tracking (FDLAT) algorithm based on the muti-domain network (MDNet). The team, lead by XU ZhiYong and ZHANG JianLin, researchers from IOE’s photoelectric detection and signal processing laboratory, propose this new algorithm and apply it to aircraft target real-time tracking in complex scenes. The algorithm can enhance the target characterization, and effectively overcome the target pose, complex scene interference and other problems. 

Figure 1. The fast deep learning for aircraft tracking (FDLAT) networks 

Now, the mainstream target tracking algorithms mainly include traditional target tracking methods and target tracking algorithms based on deep learning. Traditional target tracking methods perform well in real-time tracking, but its accuracy and robustness in different application environments are limited due to the limited feature extraction capability. The target tracking algorithms based on deep learning perform well in feature extraction, but the real-time performance is difficult to guarantee due to its complex calculation. 

The research team further optimizes the fully connected layer and regression layer in FDLAT networks to overcome these problems. As a result, the processing speed of target tracking is effectively promoted, and the accuracy of target recognition and target tracking is promoted too. After optimization, the new algorithm can stably track target in presence of attitude changes, scene interference and scale changes. 

Target tracking is an important research hotspot in the field of computer vision, with a wide range of applications, including drone surveillance, driverless, pedestrian and vehicle monitoring, etc. Target tracking has been in existence since the early 1950s. Despite the large number of research results, real-time target tracking in complex scenes is still difficult to achieve. Target deformation, fast motion and blur, illumination changes, scale changes, and occlusion in the target tracking process are still the arduous challenges for stable target tracking. 

The research, published in Opto-Electronic Engineering, 2019, 46(9), 180261, is supported by the Major Special Fund funded projects.